1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (C) 2001, 2002 Cluster File Systems, Inc.
5 * Author: Peter J. Braam <braam@clusterfs.com>
6 * Author: Phil Schwan <phil@clusterfs.com>
8 * This file is part of Lustre, http://www.lustre.org.
10 * Lustre is free software; you can redistribute it and/or
11 * modify it under the terms of version 2 of the GNU General Public
12 * License as published by the Free Software Foundation.
14 * Lustre is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 * GNU General Public License for more details.
19 * You should have received a copy of the GNU General Public License
20 * along with Lustre; if not, write to the Free Software
21 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
23 * Storage Target Handling functions
24 * Lustre Object Server Module (OST)
26 * This server is single threaded at present (but can easily be multi
27 * threaded). For testing and management it is treated as an
28 * obd_device, although it does not export a full OBD method table
29 * (the requests are coming in over the wire, so object target
30 * modules do not have a full method table.)
34 #define DEBUG_SUBSYSTEM S_OST
36 #include <linux/module.h>
37 #include <linux/obd_ost.h>
38 #include <linux/lustre_net.h>
39 #include <linux/lustre_dlm.h>
40 #include <linux/init.h>
42 static int ost_destroy(struct ptlrpc_request *req)
44 struct lustre_handle *conn = (struct lustre_handle *)req->rq_reqmsg;
45 struct ost_body *body;
46 int rc, size = sizeof(*body);
49 body = lustre_msg_buf(req->rq_reqmsg, 0);
51 rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
55 req->rq_status = obd_destroy(conn, &body->oa, NULL);
59 static int ost_getattr(struct ptlrpc_request *req)
61 struct lustre_handle *conn = (struct lustre_handle *)req->rq_reqmsg;
62 struct ost_body *body, *repbody;
63 int rc, size = sizeof(*body);
66 body = lustre_msg_buf(req->rq_reqmsg, 0);
68 rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
72 repbody = lustre_msg_buf(req->rq_repmsg, 0);
73 memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
74 req->rq_status = obd_getattr(conn, &repbody->oa, NULL);
78 static int ost_statfs(struct ptlrpc_request *req)
80 struct lustre_handle *conn = (struct lustre_handle *)req->rq_reqmsg;
81 struct obd_statfs *osfs;
83 int rc, size = sizeof(*osfs);
86 rc = obd_statfs(conn, &sfs);
88 CERROR("ost: statfs failed: rc %d\n", rc);
93 rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
97 osfs = lustre_msg_buf(req->rq_repmsg, 0);
98 memset(osfs, 0, size);
99 obd_statfs_pack(osfs, &sfs);
103 static int ost_open(struct ptlrpc_request *req)
105 struct lustre_handle *conn = (struct lustre_handle *)req->rq_reqmsg;
106 struct ost_body *body, *repbody;
107 int rc, size = sizeof(*body);
110 body = lustre_msg_buf(req->rq_reqmsg, 0);
112 rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
116 repbody = lustre_msg_buf(req->rq_repmsg, 0);
117 memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
118 req->rq_status = obd_open(conn, &repbody->oa, NULL);
122 static int ost_close(struct ptlrpc_request *req)
124 struct lustre_handle *conn = (struct lustre_handle *)req->rq_reqmsg;
125 struct ost_body *body, *repbody;
126 int rc, size = sizeof(*body);
129 body = lustre_msg_buf(req->rq_reqmsg, 0);
131 rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
135 repbody = lustre_msg_buf(req->rq_repmsg, 0);
136 memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
137 req->rq_status = obd_close(conn, &repbody->oa, NULL);
141 static int ost_create(struct ptlrpc_request *req)
143 struct lustre_handle *conn = (struct lustre_handle *)req->rq_reqmsg;
144 struct ost_body *body, *repbody;
145 int rc, size = sizeof(*body);
148 body = lustre_msg_buf(req->rq_reqmsg, 0);
150 rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
154 repbody = lustre_msg_buf(req->rq_repmsg, 0);
155 memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
156 req->rq_status = obd_create(conn, &repbody->oa, NULL);
160 static int ost_punch(struct ptlrpc_request *req)
162 struct lustre_handle *conn = (struct lustre_handle *)req->rq_reqmsg;
163 struct ost_body *body, *repbody;
164 int rc, size = sizeof(*body);
167 body = lustre_msg_buf(req->rq_reqmsg, 0);
169 rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
173 repbody = lustre_msg_buf(req->rq_repmsg, 0);
174 memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
175 req->rq_status = obd_punch(conn, &repbody->oa, NULL,
176 repbody->oa.o_blocks, repbody->oa.o_size);
180 static int ost_setattr(struct ptlrpc_request *req)
182 struct lustre_handle *conn = (struct lustre_handle *)req->rq_reqmsg;
183 struct ost_body *body, *repbody;
184 int rc, size = sizeof(*body);
187 body = lustre_msg_buf(req->rq_reqmsg, 0);
189 rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
193 repbody = lustre_msg_buf(req->rq_repmsg, 0);
194 memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
195 req->rq_status = obd_setattr(conn, &repbody->oa, NULL);
199 static int ost_brw_read(struct ptlrpc_request *req)
201 struct lustre_handle *conn = (struct lustre_handle *)req->rq_reqmsg;
202 struct ptlrpc_bulk_desc *desc;
203 void *tmp1, *tmp2, *end2;
204 struct niobuf_remote *remote_nb;
205 struct niobuf_local *local_nb = NULL;
206 struct obd_ioobj *ioo;
207 struct ost_body *body;
208 int rc, cmd, i, j, objcount, niocount, size = sizeof(*body);
211 body = lustre_msg_buf(req->rq_reqmsg, 0);
212 tmp1 = lustre_msg_buf(req->rq_reqmsg, 1);
213 tmp2 = lustre_msg_buf(req->rq_reqmsg, 2);
214 end2 = (char *)tmp2 + req->rq_reqmsg->buflens[2];
215 objcount = req->rq_reqmsg->buflens[1] / sizeof(*ioo);
216 niocount = req->rq_reqmsg->buflens[2] / sizeof(*remote_nb);
219 for (i = 0; i < objcount; i++) {
220 ost_unpack_ioo(&tmp1, &ioo);
221 if (tmp2 + ioo->ioo_bufcnt > end2) {
223 GOTO(out, rc = -EFAULT);
225 for (j = 0; j < ioo->ioo_bufcnt; j++)
226 ost_unpack_niobuf(&tmp2, &remote_nb);
229 rc = lustre_pack_msg(1, &size, NULL, &req->rq_replen, &req->rq_repmsg);
232 OBD_ALLOC(local_nb, sizeof(*local_nb) * niocount);
233 if (local_nb == NULL)
236 /* The unpackers move tmp1 and tmp2, so reset them before using */
237 tmp1 = lustre_msg_buf(req->rq_reqmsg, 1);
238 tmp2 = lustre_msg_buf(req->rq_reqmsg, 2);
239 req->rq_status = obd_preprw(cmd, conn, objcount,
240 tmp1, niocount, tmp2, local_nb, NULL);
245 desc = ptlrpc_prep_bulk(req->rq_connection);
247 GOTO(out_local, rc = -ENOMEM);
248 desc->b_portal = OST_BULK_PORTAL;
250 for (i = 0; i < niocount; i++) {
251 struct ptlrpc_bulk_page *bulk;
252 bulk = ptlrpc_prep_bulk_page(desc);
254 GOTO(out_bulk, rc = -ENOMEM);
255 remote_nb = &(((struct niobuf_remote *)tmp2)[i]);
256 bulk->b_xid = remote_nb->xid;
257 bulk->b_buf = (void *)(unsigned long)local_nb[i].addr;
258 bulk->b_buflen = PAGE_SIZE;
261 rc = ptlrpc_send_bulk(desc);
265 #warning OST must time out here.
266 wait_event(desc->b_waitq, ptlrpc_check_bulk_sent(desc));
267 if (desc->b_flags & PTL_RPC_FL_INTR)
270 /* The unpackers move tmp1 and tmp2, so reset them before using */
271 tmp1 = lustre_msg_buf(req->rq_reqmsg, 1);
272 tmp2 = lustre_msg_buf(req->rq_reqmsg, 2);
273 req->rq_status = obd_commitrw(cmd, conn, objcount,
274 tmp1, niocount, local_nb, NULL);
277 ptlrpc_free_bulk(desc);
279 OBD_FREE(local_nb, sizeof(*local_nb) * niocount);
282 ptlrpc_error(req->rq_svc, req);
284 ptlrpc_reply(req->rq_svc, req);
288 static int ost_brw_write(struct ptlrpc_request *req)
290 struct lustre_handle *conn = (struct lustre_handle *)req->rq_reqmsg;
291 struct ptlrpc_bulk_desc *desc;
292 struct niobuf_remote *remote_nb;
293 struct niobuf_local *local_nb, *lnb;
294 struct obd_ioobj *ioo;
295 struct ost_body *body;
296 int cmd, rc, i, j, objcount, niocount, size[2] = {sizeof(*body)};
297 void *tmp1, *tmp2, *end2;
298 void *desc_priv = NULL;
300 struct ptlrpc_service *srv;
304 body = lustre_msg_buf(req->rq_reqmsg, 0);
305 tmp1 = lustre_msg_buf(req->rq_reqmsg, 1);
306 tmp2 = lustre_msg_buf(req->rq_reqmsg, 2);
307 end2 = (char *)tmp2 + req->rq_reqmsg->buflens[2];
308 objcount = req->rq_reqmsg->buflens[1] / sizeof(*ioo);
309 niocount = req->rq_reqmsg->buflens[2] / sizeof(*remote_nb);
312 for (i = 0; i < objcount; i++) {
313 ost_unpack_ioo((void *)&tmp1, &ioo);
314 if (tmp2 + ioo->ioo_bufcnt > end2) {
318 for (j = 0; j < ioo->ioo_bufcnt; j++)
319 ost_unpack_niobuf((void *)&tmp2, &remote_nb);
322 size[1] = niocount * sizeof(*remote_nb);
323 rc = lustre_pack_msg(2, size, NULL, &req->rq_replen, &req->rq_repmsg);
326 remote_nb = lustre_msg_buf(req->rq_repmsg, 1);
328 OBD_ALLOC(local_nb, niocount * sizeof(*local_nb));
329 if (local_nb == NULL)
330 GOTO(out, rc = -ENOMEM);
332 /* The unpackers move tmp1 and tmp2, so reset them before using */
333 tmp1 = lustre_msg_buf(req->rq_reqmsg, 1);
334 tmp2 = lustre_msg_buf(req->rq_reqmsg, 2);
335 req->rq_status = obd_preprw(cmd, conn, objcount,
336 tmp1, niocount, tmp2, local_nb, &desc_priv);
338 GOTO(out_free, rc = 0); /* XXX is this correct? */
340 desc = ptlrpc_prep_bulk(req->rq_connection);
342 GOTO(fail_preprw, rc = -ENOMEM);
344 desc->b_portal = OSC_BULK_PORTAL;
345 desc->b_desc_private = desc_priv;
346 memcpy(&(desc->b_conn), &conn, sizeof(conn));
348 srv = req->rq_obd->u.ost.ost_service;
349 spin_lock(&srv->srv_lock);
350 xid = srv->srv_xid++; /* single xid for all pages */
351 spin_unlock(&srv->srv_lock);
353 for (i = 0, lnb = local_nb; i < niocount; i++, lnb++) {
354 struct ptlrpc_bulk_page *bulk;
356 bulk = ptlrpc_prep_bulk_page(desc);
358 GOTO(fail_bulk, rc = -ENOMEM);
360 bulk->b_xid = xid; /* single xid for all pages */
362 bulk->b_buf = lnb->addr;
363 bulk->b_page = lnb->page;
364 bulk->b_flags = lnb->flags;
365 bulk->b_dentry = lnb->dentry;
366 bulk->b_buflen = PAGE_SIZE;
369 /* this advances remote_nb */
370 ost_pack_niobuf((void **)&remote_nb, lnb->offset, lnb->len, 0,
374 rc = ptlrpc_register_bulk(desc);
379 ptlrpc_reply(req->rq_svc, req);
381 #warning OST must time out here.
382 wait_event(desc->b_waitq, desc->b_flags & PTL_BULK_FL_RCVD);
384 rc = obd_commitrw(cmd, conn, objcount, tmp1, niocount, local_nb,
385 desc->b_desc_private);
386 ptlrpc_free_bulk(desc);
389 OBD_FREE(local_nb, niocount * sizeof(*local_nb));
393 ptlrpc_error(req->rq_svc, req);
395 ptlrpc_reply(req->rq_svc, req);
400 ptlrpc_free_bulk(desc);
402 /* FIXME: how do we undo the preprw? */
406 static int ost_handle(struct ptlrpc_request *req)
411 rc = lustre_unpack_msg(req->rq_reqmsg, req->rq_reqlen);
412 if (rc || OBD_FAIL_CHECK(OBD_FAIL_OST_HANDLE_UNPACK)) {
413 CERROR("lustre_ost: Invalid request\n");
417 if (req->rq_reqmsg->type != PTL_RPC_MSG_REQUEST) {
418 CERROR("lustre_ost: wrong packet type sent %d\n",
419 req->rq_reqmsg->type);
420 GOTO(out, rc = -EINVAL);
423 if (req->rq_reqmsg->opc != OST_CONNECT &&
424 req->rq_export == NULL) {
425 CERROR("lustre_ost: operation %d on unconnected OST\n",
426 req->rq_reqmsg->opc);
427 GOTO(out, rc = -ENOTCONN);
430 if (strcmp(req->rq_obd->obd_type->typ_name, "ost") != 0)
431 GOTO(out, rc = -EINVAL);
433 switch (req->rq_reqmsg->opc) {
435 CDEBUG(D_INODE, "connect\n");
436 OBD_FAIL_RETURN(OBD_FAIL_OST_CONNECT_NET, 0);
437 rc = target_handle_connect(req);
440 CDEBUG(D_INODE, "disconnect\n");
441 OBD_FAIL_RETURN(OBD_FAIL_OST_DISCONNECT_NET, 0);
442 rc = target_handle_disconnect(req);
445 CDEBUG(D_INODE, "create\n");
446 OBD_FAIL_RETURN(OBD_FAIL_OST_CREATE_NET, 0);
447 rc = ost_create(req);
450 CDEBUG(D_INODE, "destroy\n");
451 OBD_FAIL_RETURN(OBD_FAIL_OST_DESTROY_NET, 0);
452 rc = ost_destroy(req);
455 CDEBUG(D_INODE, "getattr\n");
456 OBD_FAIL_RETURN(OBD_FAIL_OST_GETATTR_NET, 0);
457 rc = ost_getattr(req);
460 CDEBUG(D_INODE, "setattr\n");
461 OBD_FAIL_RETURN(OBD_FAIL_OST_SETATTR_NET, 0);
462 rc = ost_setattr(req);
465 CDEBUG(D_INODE, "setattr\n");
466 OBD_FAIL_RETURN(OBD_FAIL_OST_OPEN_NET, 0);
470 CDEBUG(D_INODE, "setattr\n");
471 OBD_FAIL_RETURN(OBD_FAIL_OST_CLOSE_NET, 0);
475 CDEBUG(D_INODE, "write\n");
476 OBD_FAIL_RETURN(OBD_FAIL_OST_BRW_NET, 0);
477 rc = ost_brw_write(req);
478 /* ost_brw sends its own replies */
481 CDEBUG(D_INODE, "read\n");
482 OBD_FAIL_RETURN(OBD_FAIL_OST_BRW_NET, 0);
483 rc = ost_brw_read(req);
484 /* ost_brw sends its own replies */
487 CDEBUG(D_INODE, "punch\n");
488 OBD_FAIL_RETURN(OBD_FAIL_OST_PUNCH_NET, 0);
492 CDEBUG(D_INODE, "statfs\n");
493 OBD_FAIL_RETURN(OBD_FAIL_OST_STATFS_NET, 0);
494 rc = ost_statfs(req);
497 CDEBUG(D_INODE, "enqueue\n");
498 OBD_FAIL_RETURN(OBD_FAIL_LDLM_ENQUEUE, 0);
499 rc = ldlm_handle_enqueue(req);
504 CDEBUG(D_INODE, "convert\n");
505 OBD_FAIL_RETURN(OBD_FAIL_LDLM_CONVERT, 0);
506 rc = ldlm_handle_convert(req);
511 CDEBUG(D_INODE, "cancel\n");
512 OBD_FAIL_RETURN(OBD_FAIL_LDLM_CANCEL, 0);
513 rc = ldlm_handle_cancel(req);
517 case LDLM_BL_CALLBACK:
518 case LDLM_CP_CALLBACK:
519 CDEBUG(D_INODE, "callback\n");
520 CERROR("callbacks should not happen on OST\n");
522 OBD_FAIL_RETURN(OBD_FAIL_LDLM_BL_CALLBACK, 0);
525 req->rq_status = -ENOTSUPP;
526 rc = ptlrpc_error(req->rq_svc, req);
532 //req->rq_status = rc;
534 CERROR("ost: processing error (opcode=%d): %d\n",
535 req->rq_reqmsg->opc, rc);
536 ptlrpc_error(req->rq_svc, req);
538 CDEBUG(D_INODE, "sending reply\n");
539 if (req->rq_repmsg == NULL)
540 CERROR("handler for opcode %d returned rc=0 without "
541 "creating rq_repmsg; needs to return rc != "
542 "0!\n", req->rq_reqmsg->opc);
543 ptlrpc_reply(req->rq_svc, req);
549 #define OST_NUM_THREADS 6
551 /* mount the file system (secretly) */
552 static int ost_setup(struct obd_device *obddev, obd_count len, void *buf)
554 struct obd_ioctl_data* data = buf;
555 struct ost_obd *ost = &obddev->u.ost;
556 struct obd_device *tgt;
561 if (data->ioc_inllen1 < 1) {
562 CERROR("requires a TARGET OBD UUID\n");
565 if (data->ioc_inllen1 > 37) {
566 CERROR("OBD UUID must be less than 38 characters\n");
571 tgt = class_uuid2obd(data->ioc_inlbuf1);
572 if (!tgt || !(tgt->obd_flags & OBD_ATTACHED) ||
573 !(tgt->obd_flags & OBD_SET_UP)) {
574 CERROR("device not attached or not set up (%d)\n",
576 GOTO(error_dec, err = -EINVAL);
579 err = obd_connect(&ost->ost_conn, tgt, NULL);
581 CERROR("fail to connect to device %d\n", data->ioc_dev);
582 GOTO(error_dec, err = -EINVAL);
585 ost->ost_service = ptlrpc_init_svc(64 * 1024, OST_REQUEST_PORTAL,
586 OSC_REPLY_PORTAL, "self",ost_handle);
587 if (!ost->ost_service) {
588 CERROR("failed to start service\n");
589 GOTO(error_disc, err = -EINVAL);
592 for (i = 0; i < OST_NUM_THREADS; i++) {
594 sprintf(name, "lustre_ost_%02d", i);
595 err = ptlrpc_start_thread(obddev, ost->ost_service, name);
597 CERROR("error starting thread #%d: rc %d\n", i, err);
598 GOTO(error_disc, err = -EINVAL);
605 obd_disconnect(&ost->ost_conn);
611 static int ost_cleanup(struct obd_device * obddev)
613 struct ost_obd *ost = &obddev->u.ost;
618 if ( !list_empty(&obddev->obd_exports) ) {
619 CERROR("still has clients!\n");
623 ptlrpc_stop_all_threads(ost->ost_service);
624 ptlrpc_unregister_service(ost->ost_service);
626 err = obd_disconnect(&ost->ost_conn);
628 CERROR("lustre ost: fail to disconnect device\n");
636 /* use obd ops to offer management infrastructure */
637 static struct obd_ops ost_obd_ops = {
639 o_cleanup: ost_cleanup,
642 static int __init ost_init(void)
644 class_register_type(&ost_obd_ops, LUSTRE_OST_NAME);
648 static void __exit ost_exit(void)
650 class_unregister_type(LUSTRE_OST_NAME);
653 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
654 MODULE_DESCRIPTION("Lustre Object Storage Target (OST) v0.01");
655 MODULE_LICENSE("GPL");
657 module_init(ost_init);
658 module_exit(ost_exit);